AI Moves Into the EHR at Sutter Health
The not-for-profit system integrates OpenEvidence into Epic after expanding AI across radiology and primary care
Sutter Health is embedding artificial intelligence directly inside its clinical systems following a series of pilots across imaging, documentation and screening over the past two years.
This month, the Sacramento-based not-for-profit health system announced it is integrating the evidence platform from OpenEvidence into its Epic Systems electronic health record workflows. Physicians will be able to use natural language to search clinical guidelines, peer-reviewed studies and other evidence within the EHR at the point of care, according to the companies’ announcement.
Laura Wilt, Sutter Health’s chief digital officer, said in the announcement that digital innovation “plays a central role in our work to build a more connected, proactive and sustainable healthcare system.” Dr. Ashley Beecy, Sutter’s chief AI officer, added that “Patients benefit when providers have the most current and relevant evidence incorporated into clinical decision-making.”.
The move places AI-powered evidence retrieval directly into the system clinicians already use to document visits, order tests and review results. It follows a series of AI deployments across the organization.
Scaling AI Across Clinical Workflows
Sutter’s AI strategy began with targeted use cases.
In radiology, the system partnered with Aidoc to deploy the company’s enterprise AI platform, aiOS™, across imaging workflows. Sutter and Aidoc described the collaboration as an enterprise deployment designed to deliver real-time alerts for acute findings such as pulmonary embolisms and intracranial hemorrhage inside radiologists’ existing workstations.
The health system also expanded AI-enhanced mammography screening across more than 60 imaging sites, including mobile units. Coverage described improvements in cancer detection rates and reductions in false positives associated with the AI-assisted program.
In primary care, Sutter piloted AI-enabled retinal screening for diabetic retinopathy using FDA-cleared autonomous AI technology. The program expanded beyond its initial pilot after early results, according to the American Medical Association.
Separately, Sutter partnered with Abridge to pilot ambient generative AI documentation tools that draft clinical notes from patient encounters for physician review. The goal, according to Sutter and Abridge, was to reduce documentation burden and support clinician workflow.
Those deployments show a progression: discrete clinical applications integrated into imaging and primary care workflows, followed by expansion across sites, and now integration of AI inside the enterprise EHR itself.
Embedding OpenEvidence inside Epic further moves AI from departmental tools to core clinical infrastructure. Rather than requiring clinicians to access external platforms, the evidence engine is accessible within the record system used during patient visits.
Health Systems Pair LLMs With Established Tools
Sutter’s move comes as health systems across the United States test how generative AI interacts with established clinical decision support systems.
Researchers at Mass General Brigham conducted a yearlong study comparing two large language models: GPT-4 from OpenAI and Gemini 1.5 from Google, with the health system’s traditional diagnostic decision support tool, DXplain. According to reporting on the study, the legacy system outperformed the large language models in diagnostic accuracy. The researchers concluded that combining LLMs with traditional decision support systems could improve overall performance, writing that “A hybrid approach that combines the parsing and expository linguistic capabilities of LLMs with the deterministic and explanatory capabilities of traditional DDSSs may produce synergistic benefits.”
Industry surveys from organizations such as the American Medical Association and HIMSS have reported increasing adoption of AI tools in clinical and administrative workflows over the past two years, particularly in documentation automation and imaging.
Peer-reviewed research, including large-scale trials published in The Lancet, has examined the safety and performance of AI-assisted mammography screening, reporting maintained safety thresholds alongside workflow efficiencies
Sutter’s trajectory reflects that broader pattern. The organization began with contained pilots, expanded selected tools across departments, and is now integrating AI directly into its electronic health record.
In earlier announcements about generative AI deployments, Dr. Albert Chan, Sutter’s chief health information officer, said the tools helped providers “recharge”.